This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.

Jump in the discussion.
No email address required.
Notes -
LLMs are actually great at picking up subtle nuances in the training data which humans would call stereotypes.
For chatbots, the general goal for RLHF is something like "humble, helpful, harmless". Picking a face of the Shoggoth comes with baggage. If I imagine a friendly, qualified operator of some helpline, the stereotype says they are likely a young college-educated woman.
I am sure that if you used RLHF to select a Shoggoth-face which talked like 4chan, you would get very different politics along with it.
This seems utterly pathetic on the part of the AI companies.
I mean, there are things which are outrage bait which need to be fixed, if your LLM happily generates python code which prints a recipe for methamphetamine, insults racial minorities with the right prompt or the like, you want to CYA and claim that you fixed this as soon as you became aware of the issue, even if the fix is just filtering any queries which mention both meth and python.
This however is not outrage bait. Papering over the flaws of your model one by one is utterly pointless.
Suppose I have a colleague who has written a function, and I tell him that actually his function fails for x=-1/3. What he then does is to add the following:
This might be an acceptable short-term stopgap if I had indicated that his function failing for x=-1./3 (and only that value) is a showstopper for me, but in general it is a terrible idea. If his code fails for every x==-1./prime, then his approach will not converge to correct code any time soon, instead, he is simply papering over the specific errors people have found, e.g. cheating to make his code look correct when he knows it is not.
Ideally, AI companies would indicate a minimum of trustworthiness given that they are in a position of power (and possibly going to summon the demon which wipes out mankind). If they are already willing to cheat with this stuff which nobody would care about, what does that promise for the more serious stuff? It is like watching your new chief of police using a length of wire to steal a can of soda from a vending machine.
--
With regard to the content of the mail, this is a trick question of a similar category as the question if the pool of the Titanic is full or not. A human who thinks about these will typically think in images rather than words. I think that the Titanic one is a bit more vicious because the behavior of fluids is rarely spelled out by humans, while 'I drove my car through the car wash' should appear in the training data, while seawater flooding the swimming pool of a sinking vessel might not have appeared explicitly at all.
This is an old criticism, older than me and I'm old as all hell. And indeed, it would make sense without realizing that training always has a constraint (implicit or explicit) on the size of the function, or in the case of LLMs, the number of parameters.
More options
Context Copy link
More options
Context Copy link